Research On Computer Aided Automatic Sex Determination Of Skull

The sex determination of the skull is one of the hot research topics in forensic anthropology.It has important research value in the field of criminal investigation and archaeological anthropology.The traditional morphological methods and measurement methods rely on the subjective experience of experts,high measurement accuracy requirements and cumbersome operations,which lead to large errors in gender identification and complex and time-consuming identification process.The computer-assisted gender identification of the skull not only effectively shortens the period of gender identification of the skull,but also avoids secondary damage to the skull during measurement,and the recognition rate is high.Therefore,three-dimensional skull as the object of study,combined with image processing technology and machine learning methods for skull sex classification research has become a hot spot.Traditional morphological methods and measurement methods rely on the subjective experience of experts,require high measurement accuracy and complicated operation,resulting in large errors in gender identification and complex identification process time-consuming.In order to solve the problem of traditional methods,this paper proposes two automatic methods of skull sex identification to achieve three-dimensional skull sex identification.The main research works of this paper are included as follows:(1)A method for sex determination of skulls based on statistical deformation model and support vector machine is proposed.Firstly,in order to establish a statistical deformation model scientifically and effectively,it is necessary to establish a point correspondence relationship for all skull samples,use the TPS algorithm to deform the global non-rigid coarse registration,and then use the ICP algorithm for fine registration,and finally achieve all the skull registration,establish Their point correspondence.Then,the statistical deformation model of skull is established by PCA and the model parameters are solved to get the skull eigenvectors.Finally,the dimension of the extracted eigenvectors is reduced by SVM and the classifier is designed to realize the skull gender classification.The experimental results show that the statistical deformation model can characterize the skull more effectively and achieve higher accuracy.Compared with previous methods,this method does not measure distance or volume dependent variables,but describes the global shape of the skull.When the statistical shape model and discriminant function are established,it is easy to determine the unknown skull sex.It does not need professional knowledge and tedious manual measurement,but alsocan ensure a higher accuracy.(2)A method of sex identification of skull based on improved convolution neural network and least square method is proposed.Firstly,the skull multi-angle images are acquired and the skull training sample set is obtained.Then,the improved CNN is used to obtain the skull features and calculate the probability that the skull images belong to men or women respectively,and the content information of the images is mined in depth.Finally,the optimal parameters are obtained by feature fusion using the least square method,and then the classification model is constructed.Finally,the gender identification of intact skull and non-intact skull is realized and the recognition accuracy is high.Experiments show that this method can get better classification performance.Obtaining skull features from multi-angle skull images improves the influence of the posterior part of the skull on dimorphism,effectively reduces the errors caused by human vision and subjective factors,and also reduces the time consumption and tedious preprocessing.Compared with the traditional machine learning method,the depth learning algorithm is more suitable for large sample data sets,and it has strong learning ability,high classification accuracy,strong robustness and fault tolerance to noise data.